ITPP: Learning Disentangled Event Dynamics in Marked Temporal Point Processes
This work addresses the challenge of disentangling event dynamics for researchers and practitioners in temporal modeling, offering a novel method to improve performance in applications like event prediction, though it is incremental in advancing MTPP architectures.
The paper tackles the problem of modeling asynchronous event sequences in Marked Temporal Point Processes (MTPPs) by addressing the entanglement of event type information in existing models, which leads to performance degradation and overfitting. The result is ITPP, a channel-independent architecture that consistently outperforms state-of-the-art MTPP models in predictive accuracy and generalization across multiple datasets.
Marked Temporal Point Processes (MTPPs) provide a principled framework for modeling asynchronous event sequences by conditioning on the history of past events. However, most existing MTPP models rely on channel-mixing strategies that encode information from different event types into a single, fixed-size latent representation. This entanglement can obscure type-specific dynamics, leading to performance degradation and increased risk of overfitting. In this work, we introduce ITPP, a novel channel-independent architecture for MTPP modeling that decouples event type information using an encoder-decoder framework with an ODE-based backbone. Central to ITPP is a type-aware inverted self-attention mechanism, designed to explicitly model inter-channel correlations among heterogeneous event types. This architecture enhances effectiveness and robustness while reducing overfitting. Comprehensive experiments on multiple real-world and synthetic datasets demonstrate that ITPP consistently outperforms state-of-the-art MTPP models in both predictive accuracy and generalization.